Open Stack

On Oct 24, 2016, at 6:06 PM, Vaidyanath Manogaran wrote:

> I have my controller node and compute01 node in one vlan(118)
> I have another compute node in different vlan(112)
> 
> Can I have the dhcp client running in both the compute nodes instead of the controller node?
> 
> Or Is it mandatory that the management network should be in the same network?

You need to understand that there's a [minor] difference between "a control
node" and a "network node".

I do, as far as I can understand from what you've said so far, do it like you do - I
have two control nodes (replicated and loadbalanced) and fourteen compute
nodes.

The "network node" is 'running on' (as in, it's the same as) the "control node".

The "network node" is the part that takes care of all the networking stuff (doh! :).
Pretty much, the part that sits between the physical network and all of the
tenant networks (where your virtual machines run).


The way I designed _my_ system, is that I have one external network (which
is basically the "management network") and a couple of tenant networks. Each
tenant network have at least one subnet.

And this is where dnsmasq runs on - the subnet. So there's one dnsmasq process
per subnet. NOT on the physical network.. Which is why you can't have a central
DHCP server (because that will run on the _physical_ network - somewhere).


I know, from experience, that Neutron is the biggest obstacle to learn - it's _HUGE_!
But you really need to go back to the documentation (as they are) and study them
in more detail. Especially the part where they [try to] explain the architecture..
--
Imagine you're an idiot and then imagine you're in
the government. Oh, sorry. Now I'm repeating myself
- Mark Twain

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20161024/c2b25bd6/attachment.html>